Biostatistics and Research Methodology
Core concepts
Candidates must distinguish data types and scales, descriptive statistics, sampling distributions, confidence intervals, hypothesis tests, type I and II errors, power, and sample-size determinants. Know when to use t-tests, chi-square, ANOVA, non-parametric tests, correlation, and regression; understand confounding versus mediation, interaction, and adjustment. For trials, grasp randomization, allocation concealment, blinding, intention-to-treat, per-protocol analysis, loss to follow-up, and multiple comparisons. For diagnostic studies, know sensitivity, specificity, predictive values, likelihood ratios, and reliability measures such as kappa.
Traps
A p-value is not the probability that the null hypothesis is true, and a 95% confidence interval is not a 95% probability that the parameter lies within that particular interval. Statistical significance is not clinical importance. Watch for standard deviation versus standard error, one-tailed tests used without justification, changed primary outcomes, per-protocol claims from a trial analysed as treated, confounding by indication, regression to the mean, and subgroup fishing.
Worked example
A test has 90% sensitivity and 95% specificity; prevalence is 1%. In 10,000 people: 100 have disease, 90 are true positives; 9,900 are healthy, and 5% (495) are false positives. Positive predictive value = 90/585 = 15.4%. The same test in a 20% prevalence clinic population gives PPV = 180/(180+400) = 31%. PPV depends on prevalence; sensitivity and specificity do not.
Rule of thumb
Report absolute effects with confidence intervals, not just p-values, and match the analysis to the design.